International Journal of Technical Research and Applications e-ISSN: 2320-8163,

www.ijtra.com Volume 3, Issue 2 (Mar-Apr 2015), PP. 07-11

FAULT DETECTION AND DIAGNOSIS IN

CONTINUOUS STIRRED TANK REACTOR
(CSTR)
R. Gowthami, Dr. S. Vijayachitra
Department of EIE
Kongu Engineering College
Erode, India
gowthami.ei@gmail.com, dr.svijayachitra@gmail.com

dCa
F
Cai Ca K 0 e RT Ca

dt
V

ra K 0 e

E
RT

(1)

Cai = Inlet reactant concentration in reactor

Ca

= Outlet reactant concentration in reactor

=Inlet feed flow rate

=Volume of reactor

K0

=Frequency factor

=Activation energy

=Universal gas constant

=Reactor temperature
2
1.8

I. INTRODUCTION
The Continuous Stirred Tank Reactor (CSTR) system is
highly nonlinear, exothermic, irreversible first order process.
In CSTR, when the reactants are added into the tank the stirrer
will stir the reactants to give desired product. Once the
equipment is running, it is usually operated at steady state and
designed to achieve well mixing. The CSTR also known as
back mix reactor is commonly used as perfect reactor type in
chemical engineering. A CSTR is a model which is used to
estimate individual operation variables while using a
continuous agitated tank reactor in order to obtain the required
output. CSTR is widely used in the organic chemicals industry
for medium and large scale production. The reactor is widely
operated by three control loops that will regulate the outlet
temperature and the inlet flow rate of the reactant tank level.
During the process the heat will be generated and hence the
heat of reaction can be removed by a coolant medium that
flows through a jacket around the reactor. During the CSTR
process the faults occur which further leads to inaccurate
result. A fault is defined as an unexpected change of the
system functionality which may be related to a failure in a
physical component such as sensor and actuator. In order to
detect various types of faults in CSTR the Neural Network
Predictive Controller can be applied.

II. MATHEMATICAL MODELING OF CSTR

A reaction will create new components while
simultaneously reducing reactant concentrations. The reaction
may give off heat or give required energy to progress the
process [2]. Figure 1. shows the CSTR consists of one inlet
reactants and one outlet flow rate. In CSTR the outlet
concentration is controlled by changing the volumetric flow
rate [8]. Figure 2. shows the characteristic of the concentration
and volumetric flow rate. Figure 3. shows the servo response
of Continuous Stirred Tank Reactor. If the outlet concentration
is not equal to the set point then the volumetric flow rate is
adjusted in order to make outlet concentration to reach the
desired set point.
By the mass balance (typical unit, Kg/s),
The Rate of change of mass in the system=The Rate of mass
flow in Rate of mass flow out

AbstractContinuous Stirred Tank Reactor (CSTR) here is

considered as a nonlinear process. The CSTR is widely used in
many chemical plants. Due to changes in process parameters the
accuracy of final product can be reduced. In order to get accurate
final product the faults developed in CSTR during the chemical
reaction need to be diagnosed. If not, the faults may lead to
degrade the performance of the system. For this purpose there
are various fault diagnosis methods are to be considered. Among
the methods, the neural network predictive controller can be used
to detect faults in CSTR. Servo response is performed to
understand the behavior of CSTR. By detecting various faults
and with suitable control techniques, the accuracy of the
desirable products in CSTR can be improved.
Index TermsContinuous Stirred Tank Reactor, Neural
Network Predictive Controller, Fuzzy Logic Controller, Fault
detection and estimation.

1.6

volumetric flow of coolant

1.4

Figure 1. Continuous Stirred Tank Reactor

10
Time

12

14

16

18

20

10
time

12

14

16

18

20

21
20.5
20
19.5
19

Figure 2. Behavior of Manipulated Variable- Controller Variable in

Continuous Stirred Tank Reactor

7|P age

volumetric flow of coolant

International Journal of Technical Research and Applications e-ISSN: 2320-8163,

www.ijtra.com Volume 3, Issue 2 (Mar-Apr 2015), PP. 07-11
The fuzzy logic controller uses inaccurate or not precise
2
input and output variables by using membership function. The
1.5
membership function can be expressed in terms of linguistic
1
variables. In the fuzzy logic controller the if-then rules are
0.5
used to identify the fault and fault free data. At the same time
0
0
10
20
30
40
50
60
Time
it provide the symptoms for faulty and fault less data by
20
predefined fuzzy sets. The amount of fault that present is
15
obtained by comparing the rules of reference model and with
10
the rules of fuzzy reference model which is identified by using
5
data obtained from CSTR plant [2],[3]. Here the process
0
0
10
20
30
40
50
60
historical data is used [7]. The process historical data means
time
data obtained from CSTR plant.
Figure 3. Servo Response of Continuous Stirred Tank Reactor
IV. FUZZY DECISION MAKING
III. METHOD OF FAULT DIAGNOSIS
The fuzzy logic controller is used for direct integration of
The term fault is generally defined as a departure from an
the human operator into the fault detection and supervision of
acceptance range of an observed variable [6], [7]. Figure 4.
CSTR process. Changes in any variable in the process will
shows that the faults developed in the plant diagnosed by the
reduce the accuracy of the final product. It also provides
sensor and the sensor value will then be compared with the set
supervision of entire process. In the supervision of the process
point in the comparator. If there is any error in the sensor then
the alarm will be provided. If any error occurs in the process
the feedback controller generates a signal to the actuator. The
the alarm will be generated. For this purpose the fuzzy
actuator signal will be given to the diagnosis system. If any
decision making is necessary. In order to avoid any incorrect
error occurs in the actuator, the diagnosis system will provide
decision which may cause false alarm then the fuzzy decision
an alarm. The fault diagnosis system provides the supervision
making is used to decide whether the fault has occurred and at
system with information about the onset, location and severity
which place in the process the fault has occurred. The fuzzy
of the faults. Based on the system inputs and outputs together
decision making is similar to the expert system and
it will make fault decision information from the fault diagnosis
supervisory control.
system.
V. FAULT ESTIMATION BY FUZZY LOGIC CONTROLLER
The fuzzy logic system handles the imprecision of input
and output variables directly by defining them with
memberships and sets that can be expressed in linguistic
terms. The fuzzy reference models are made up from if-then
rules which describe the symptoms of faulty and fault free of
predefined fuzzy reference sets. A particular model is defined
by specifying the values of the elements of its associated fuzzy
relational array. Each element of the array is a measure of the
credibility that the associated rules correctly describes the
behavior of the system around the particular operating point.
For fault detection and estimation in CSTR using Fuzzy Logic
Figure 4. General Components of Fault Diagnosis Framework
Controller. The fault level in CSTR depends on four variables
such as feed flow rate (F0), outlet reactant concentration (CA0),
FUZZY LOGIC CONTROLLER
coolant temperature (TC0) and feed temperature (T 0). These
Fuzzy logic is an extension of crisp set. A fuzzy logic
four variables fault level depends on the volume of reactor
system is a type of control system that work based on fuzzy
(V), reactant concentration in the reactor (CA), reactor
logic. The fuzzy logic controller is based on mathematical
temperature (T) and Jacket temperature (T C) [5]. Based on
system that evaluates input values for fuzzy logic system in
these fault range the Fuzzy Logic Controller is developed. The
terms of sensible variables. The logical variables take values
Fuzzy Logic Controller consists of four inputs and four
in two ways they are continuous and discrete values. If the
outputs. The membership function is developed for those four
logical variable uses continuous value which ranges from 0 to
inputs and four outputs. The rules are created for these
1 in contrast to digital logic and if logical variables for fuzzy
variables. Based on the input fault range the fault level is
logic operate on discrete values which is either 1 or 0 that is
estimated by using Fuzzy Logic Controller.
true or false respectively. Fuzzy system is useful in any
The figure.5 shows various inputs to the Fuzzy Inference
situation in which the measurement depends strongly on the
System such as volume of reactor (V), reactant concentration
context or human opinion. A fuzzy reasoning algorithm
in the reactor (CA), reactor temperature (T) and jacket
establishes a preliminary fuzzy diagnosis system.
temperature (TC) and the fault outputs considered for Fuzzy
Fuzzy logic controller is commonly used in machine
Inference System are feed flow rate (F0), outlet reactant
control operations. When the logic involved in fuzzy logic
concentration (CA0), coolant temperature (TC0) and feed
controller can deal with concepts which can be expressed in
temperature (T0).
terms as neither true nor false but considerably as partially
true. In such cases the rules are created to deal both with true,
partially true, false and partially false. Fuzzy sets mean the
input variables. The input variables in a fuzzy operating
machine can be obtained by mapping the sets of membership
functions. Fuzzificaion means the series of action for
converting a crisp value to a fuzzy value.
Figure 5. Fuzzy Inference System Editor for overall fault

8|P age

International Journal of Technical Research and Applications e-ISSN: 2320-8163,

www.ijtra.com Volume 3, Issue 2 (Mar-Apr 2015), PP. 07-11
With the help of Fuzzy Logic Controller, various faults
considered are detected and estimated and the results obtained
are shown in Table 1
Table 1: Fuzzy Logic Based Fault Estimation in CSTR

Figure 8. Fuzzy Inference System Editor for fault in coolant temperature

(TC0)
Table 4: Fuzzy Logic Based Fault Estimation for Coolant Temperature
(TC0)

Similarly the fault level for individual variable can also be

obtained. Figure.6 shows the fault in feed flow rate (F 0)
depends on volume of reactor (V), reactant concentration in
the reactor (CA), reactor temperature (T) and jacket
temperature (TC). After developing the fuzzy logic controller
the estimation of fault level in feed flow rate (F0) can be done
is shown in table 2.
Similarly the fault level in feed temperature (T 0) depends
on volume of reactor (V), reactant concentration in the reactor
(CA), reactor temperature (T) and jacket temperature (T C) as
shown in the figure 9. Table 5 shows the estimation of fault
level in feed temperature (T 0).
Figure 6. Fuzzy Inference System Editor for fault in feed flow rate (F 0)
Table 2: Fuzzy Logic Based Fault Estimation for Feed Flow Rate

Similarly the fault level in outlet reactant concentration

(CA0) is based on volume of reactor (V), reactant concentration
in the reactor (CA), reactor temperature (T) and jacket
temperature (TC) as shown in the figure 7. Table 3 shows the
estimation of fault level in outlet reactant concentration (C A0).

Figure 7. Fuzzy Inference System Editor for fault in outlet reactant

concentration (CA0)
Table 3: Fuzzy Logic Based Fault Estimation for Outlet Reactant
Concentration

Similarly the fault level in coolant temperature (T C0)

depends on volume of reactor (V), reactant concentration in
the reactor (CA), reactor temperature (T) and jacket
temperature (TC) as shown in the figure.8. After developing
the fuzzy logic controller the estimation of fault level in
coolant temperature (TC0) can be obtained as shown in table 4.

Figure 9. Fuzzy Inference System Editor for fault in feed temperature

(T0)
Table.5 Fuzzy Logic Based Fault Estimation for feed temperature (T 0)

VI. NEURO PREDICTIVE XL SOFTWARE

Prediction means estimate, forecast and predict the future
condition for this purpose the neural predictive XL software is
used. In Microsoft excel the predictions can be done by a built
in tool but the correctness of its results is greatly decreased
when non-linear relationship occur or if any missing data are
present. Neural networks are a tried, tested and most
commonly used technology for most complex prediction
operation. It is modeled by considering the human brain. The
neural network is joined onto one another networks of separate
processors [1], [4]. By changing the connections of these
neural networks the finding answer to a problem can be
identified. One of the major reasons that the analysts not using
these latest methods such as neural networks in order to
improve forecasts because the neural network method is very
difficult to monitor. Neuro XL Predictor removes the mental
process and practical limit by hiding the complicated nature of
its latest neural network based methods while taking profit of
analysts having present information in mind. Since users make
the act of predicting the future through the commonly seen
9|P age

International Journal of Technical Research and Applications e-ISSN: 2320-8163,

www.ijtra.com Volume 3, Issue 2 (Mar-Apr 2015), PP. 07-11
excel point of interaction, it reduces the learning time, greatly
off heat or give required energy to proceeds the process.
decreasing the intervening period between loading the
Figure 10. shows the CSTR consists of two inlet reactants and
software and performing useful predictions. The application is
one outlet flow rate. The two inlet reactants are W 1 which is
extremely known automatically and not difficult to use this
inlet flow rate of concentrated feed and W2 is inlet flow rate of
neuro XL predictor software. In neural predictive XL by the
diluted feed [2], [7].
knowledge of previous inputs and outputs it is able to predict
By the mass balance (typical unit, Kg/s),
what the future output for present input can be identified. In
The Rate of change of mass in the system=The Rate of mass
this neural predictive XL the sensor fault input and output data
flow in Rate of mass flow out
is loaded in the excel sheet by using neural predictive XL
dh
software the future output is predicted. Table.6 the fault in the
(1)
W W W
1
2
0
output can be predicted by using neuro predictor XL. In
dt
Continuous Stirred Tank Reactor the sensor fault is
(2)
W CV h
0
considered. Neuro predictor XL is a spreadsheet which is used
dh
to detect fault in Continuous Stirred Tank Reactor. The sensor
(3)
W W (CV h )
1
2
dt
fault in CSTR is used [9]. In the sensor fault the flow rate of
the liquid at the inlet ( Fi ), control valve opening
Where CV =0.2
( Fcv ), temperature of the inlet reactant ( Ti), control valve
dh
(4)
W W (0.2 h )
opening ( L ), temperature of the tank ( T ), temperature of
1
2
dt
cv
0
the outlet coolant ( T j 0), flow rate of coolant ( F j ), liquid
Where
level ( L ) are the variables considered in CSTR. By loading
h =Liquid level (cm)
the sensor faults of CSTR in the spread sheet the neuro
predictor XL software is used to predict future fault in CSTR.
W =Inlet flow rate of the concentrated feed (cm3 /s)
The neuro predictor XL software predict future fault in CSTR
1
by seeing present and past faults in sensor of CSTR. So this
W =Inlet flow rate of the diluted feed (cm3 /s)
2
neuro predictor XL software is useful for analysts to predict
future fault in CSTR. By predicting the future fault the
W =Outlet flow rate (cm3 /s)
0
analysts can take some preventive measure to reduce the fault
The concentration of the continuous stirred tank reactor is
before its effect felt in the process. By doing these measure the
given in the following equation
final outcome of the process can be maintained.
Table.6 Prediction of Fault in CSTR
dC
W
W
b (C C )( 1 ) (C C )( 2 ) r
b1
b h
b2
b h
b
dt
(5)
Where,
k C
(6)
1 b
r
b
2
(1 k C )
2 b
Where,
=The concentration of product at the output of the
C

process ( mol/cm )

Where,

Predicted
Output
( cm 3 / s )

Fi

= Flow rate of liquid at inlet

Fcv

= Control valve opening

Ti

= Temperature of the inlet reactant(C)

Lcv

= Control valve opening

T0

= Temperature of the tank (C)

Tj0

= Temperature of the outlet coolant

water (C)

Fj
L

b1

=Inlet concentration feed ( mol/cm )

3
=Outlet concentration feed ( mol/cm )
b2
W =Inlet flow rate of the concentrated feed ( cm 3 /s )
1
W =Outlet flow rate of the diluted feed ( cm 3 /s )
2
k , k =The constants associate with the rate of consumption
1 2
h =Liquid level (cm)

= Flow rate of coolant ( cm 3 / s )

= Liquid level (cm)

VII. NEURAL NETWORK PREDICTIVE CONTROLLER

The Mathematical Model of CSTR for two tank system is
A reaction will create new components while simultaneously
reducing reactant concentrations [9]. The reaction may give

Figure 10.Continuous Stirred Tank Reactor

10 | P a g e

International Journal of Technical Research and Applications e-ISSN: 2320-8163,

www.ijtra.com Volume 3, Issue 2 (Mar-Apr 2015), PP. 07-11
Servo Response of CSTR
23
Plant Output
Reference Signal
Concentration in mol/(cm)

22.5

22

21.5

21

20.5

20

Figure 11. Neural Network Predictive Controller

The neural network predictive controller is a graphical

user interface tool as shown in figure 11. From that neural
network predictive controller select the plant identification
which consists of neural network plant model. The plant
model predicts the future plant outputs. The neural network
plant model consists of one hidden layer. By generating
training data by giving series of random reference signal the
plant input and output data will be generated. By using trainlm
the plant model training is done from this the training data for
NN predictive controller will be obtained as shown in figure
12. and Validation data for NN predictive controller will be
generated as shown in figure 13.
Input
23

22
2
21
1
20
500

1000

500

Error

1000

NN Output

0.15

23

0.1
22
0.05
21
0
-0.05

20
0

500
time (s)

1000

500
time (s)

1000

Figure 12. Training Data for NN Predictive Controller

Input

Pla nt Output

23

3
22
2
21
1
0

20
0

100

200

300

Error

100

200

300

NN Output

0.04

23

0.02

22

21

-0.02

20
0

100

40

60
Time in Secs

80

100

120

Figure 14. Servo Response of CSTR

VIII. CONCLUSION AND FUTURE SCOPE

The application of Continuous Stirred Tank Reactor is
very essential nowadays. Using material balance condition, the
CSTR process is effectively modeled which provides a better
understanding of its characteristic behavior of the process. The
modeling is effectively verified by its servo response. Various
sensor faults considered in CSTR are detected and estimated in
percentage by means of using Neural Network Predictive
Controller and Fuzzy Logic Controller and found good results.
This fault detection and estimation work can be further
extended to actuator faults also.
IX. REFERENCES

20

Plant Output

200
300
time (s)

100

200
300
time (s)

Figure 13. Validation Data for NN Predictive Controller

Figure 14. shows the servo response of CSTR. By using the

neural network predictive controller plant model the
characteristic of concentration in CSTR is obtained.

[1] Dong-Juan Li., 2014, Neural network control for a class of

continuous stirred tank reactor process with dead-zone input,
Elsevier Journal of Neurocomputing, Vol.131, pp.453-459.
[2] Kanse Nitin G., Dhanke P.B. and Thombare Abhijit ., 2012,
Modeling and Simulation Study for Complex Reaction by
Using Polymath, Research Journal of Chemical Sciences,
Vol.2, No.4, pp.79-85.
[3] Manikandan P., Geetha M., Jubi K., Jovitha Jerome.,2013,
Fault Tolerant Fuzzy Gain Scheduling Proportional-IntegralDerivative Controller for Continuous Stirred Tank Reactor,
Australian Journal of Basic and Applied Sciences, Vol.7,
No.13, pp.84-93.
[4] Ribhan Zafira Abdul Rahman, Azura Che Soh, Noor Fadzlina
binti Muhammad.,2010, Fault Detection and Diagnosis for
Continuous Stirred Tank Reactor Using Neural Network,
Kathmandu University Journal of Science, Engineering and
Technology, Vol.5, No.2, pp. 66-74.
[5] Souragh
Dash,
Raghunathan
Rengaswamy,
Venkat
Venkatasubramanian.,2003, Fuzzy- logic based trend
classification for fault diagnosis of chemical processes, Elsevier
Journal of Computers and Chemical Engineering, Vol.27,
pp.347-362.
[6] Venkat Venkatasubramanian, Raghunathan Rengaswamy,
Kewen Yin, Surya N. Kavuri., 2003, A review of process fault
detection and diagnosis Part I:Quantitative model-based
methods, Elsevier Journal of Computers and Chemical
Engineering, Vol.27, pp.293-311.
[7] Venkat Venkatasubramanian, Raghunathan Rengaswamy,
Surya N. Kavuri, Kewen Yin., 2003, A review of process fault
detection and diagnosis Part III: Process history based methods,
Elsevier Journal of Computers and Chemical Engineering,
Vol.27, pp.327-346.
[8] Vishal Vishnoi, Subhransu Padhee, Gagandeep Kaur., 2012,
Controller Performance Evaluation for Concentration of
Isothermal Continuous Stirred Tank Reactor, International
Journal of Scientific and Research Publication, Vol.2, No.6,
pp.2250-3153.
[9] Yunosuke Maki and Kenneth A. Loparo.,1997, A NeuralNetwork Approach to Fault Detection and Diagnosis in
Industrial Processes, IEEE Transactions on Control Systems
Technology, Vol.5, No.6, pp.529-541.

11 | P a g e